P R A KR I T P H ON O L OG Y

A BRIEF INTRODUCTION

Introduction

The vast majority of Prakrit words are what Prakrit grammarians call tadbhavaḥ, “derived
from Sanskrit,” which means that we can specify a number of transformations which, when
applied to a given Sanskrit form, will yield the Prakrit form under consideration. In other
words, the grammar of Prakrit can largely be specified as an “extension” of the grammar of
Sanskrit using transformational rules.

To some degree this is a presentational choice. Grammarians chose to relate Prakrit to

Sanskrit because the grammar of Sanskrit was already well-known. They could have
described Prakrit “from the ground up,” without presupposing the corresponding Sanskrit
forms, as some modern pedagogues do. They could also, in principle, have used another
language, such as Pāli, as the basis for their description of Prakrit.

There is, however, an obvious logic to using Sanskrit as the basis for teaching Prakrit, which
is that Prakrit can be considered to represent a phase of linguistic development subsequent to
the phase represented by Sanskrit. This does not mean that Prakrit developed from Sanskrit,
at least if we understand these terms to refer to the languages described by Vararuci and
Pāṇini, respectively. It is accurate, however, to say that Prakrit, and the rest of the Middle
Indic languages, presuppose as their developmental basis an “Old Indic” language that was
very similar to Sanskrit, although not identical to it in all of its particulars.

Some modern scholars, like Eivind Kahrs, have maintained that the very notion of language
changing in time would have been unintelligible to premodern thinkers. I disagree. But Kahrs
was responding to a widespread conflation between the descriptive technique of deriving
Prakrit forms from Sanskrit forms, which does not imply any historical relationship between
the two sets of forms, and the historical claim that Prakrit forms arose from forms very
similar to those we encounter in Sanskrit. The reason for this conflation is that the
transformational rules connecting Sanskrit and Prakrit that were offered by the premodern
grammarians are very similar to the rules of historical sound change that were (and continue
to be) offered by philologists.

This brief introduction to Prakrit phonology presupposes Sanskrit as a basis, as the

premodern grammarians do, but it motivates the various transformations by collecting them
under the heading of several phonological constraints and the repair strategies that are
followed when those constraints are violated. Constraints are systematic patterns of
avoidance that characterize a language. Constraints are very general, and the effects of a
given constraint can be observed in many of the world’s languages. Repair strategies are
ways of “fixing” a form that runs afoul of a given constraint, usually by changing or getting
rid of a sound. It is important to note that a single constraint can often be satisfied by multiple
repair strategies, which results in doublets.

For example, English doesn’t generally allow a stop consonant to be followed by

an occlusive (including stops, affricates, and nasals) at the beginning of a word. Hence, in
loan-words from languages that allow this combination of consonants, such as Greek or

||1||
 German, we generally follow the repair strategy of deleting the first consonant, resulting in
the non-pronunciation of p in words like psychology, pneumatic, and so on.

A lot of the constraints of Prakrit phonology, and hence the motivation for many of the
sound-changes particular to Prakrit, relate to syllable structure. The same is true of
English: ŋ (= ṅ) cannot occur at the beginning of a syllable, h cannot occur at the end of a
syllable, and so on.

Note that I write long vowels (ā, ī, ū, ē and ō) in this outline unless those vowels have been
shortened in a word-internal context. But the reader must keep in mind that any long vowel
must be shortened when followed by a double consonant, even if that double consonant
occurs at the beginning of the following word.

S Y LL AB LE S TR UC T UR E
Phonemes

§1. Basic concepts.

No superheavy syllables
The syllable is a basic phonological constituent, consisting of a nucleus, which is always a
vowel. (Vowels are, in fact, sometimes defined by their syllabicity, or the fact that they can
constitute the nucleus of a syllabus.) An optional onset of one or more consonants may
precede the nucleus, and an optional coda of one or more consonants may follow it.
(Sometimes the nucleus and coda are together referred to as the rime.) Hence we can say σ =
C*VC*, or “a syllable is zero or more consonants, followed by one vowel, followed by zero
or more consonants.”

A syllable with a coda consonant is called closed, and a syllable without a coda consonant is
called open.

Syllables are generally parsed by maximizing onsets, that is, we preferentially assign
consonants to onsets rather than codas. Hence, if we represent the boundary between
syllables with a dot (.), we would parse the following Sanskrit words into syllables in the
following way:

dat.taḥ sid.dham bhak.taḥ can.da.naḥ can.draḥ kṛt.snam put.traḥ

Clusters of consonants with r are somewhat tricky—you might expect a parse like put.raḥ—
but Sanskrit grammarians treat these clusters as if they involve a doubling of the stop
consonant, which entails taking the consonant with both the preceding and following syllable,
as in the above parse of put.traḥ (see 8.4.46–52).

A fundamental feature of any syllabic is its weight, which is a function of its moras. A mora
(mātrā in Sanskrit) is an abstract phonological unit which is associated with certain
constituents of the syllable: in Sanskrit and Prakrit, for all intents and purposes, only
segments in the rime of the syllable (i.e., the vowel in the nucleus, and any consonants that
follow in the coda) are associated with moras.

Short vowels are associated with one mora. Long vowels are associated with two. The “long
diphthongs” of Sanskrit, ai and au, are associated with three, since they underlyingly consist
||2||
 of a long vowel (ā) followed by another vowel (i/ī or u/ū). Each consonant of the coda is
associated with one mora.

When we add all of the moras of a syllable together, we get its weight. Weight is a relevant
feature for many phonological and morphological processes. These processes generally treat
weight as binary, that is, they distinguish between light and heavy syllables: the former have
a one mora, and the latter have more than one. Hence, for the purposes of these processes, we
generally stop counting moras once we reach two. For example, the augment tuK is only
inserted before the suffix LyaP when the suffix follows a light syllable (hence apa-nī-ya,
after the heavy syllable nī, but vi-ji-t-ya, after the light syllable ji).

Prakrit phonology, however, makes a ternary distinction of weight:

 light syllables are those that contain one mora, namely:

o those that have a short vowel without any following consonants (C*V);
 heavy syllables are those that contain two moras, namely:
o those that have a long vowel without any following consonants (C*VV), and
o those that have a short vowel followed by a single consonants (C*VC);
 superheavy syllables are those that contain three (or theoretically more than three)
moras, namely:
o those that have a long vowel followed by one or more consonants (C*VVC+),
and
o those that have a short vowel followed by two or more consonants
(C*VC{2}).

We will see examples of all of these syllable types in the following section.

§2. No superheavy syllables.

Basic concepts
The most important constraint of Prakrit phonology, which is shared with all other Middle
Indic languages, is that no superheavy syllables are allowed. If a syllable has three or more
moras, it must be modified in some way to reduce the total number of moras to two. The
repair strategies invoked when this constraint is violated all involve the deletion of at least
one moraic segment, whether it is a part of a vowel, or a coda consonant.

§2.1.Shortening vowels
Long vowels are always shortened before a consonant in the same syllable. This can be
thought of as “shaving off” one mora of the vowel. Note that the vowels ē and ō, although
“naturally” long, can be shortened to e and o before a consonant. Some scribes, however,
write these short vowels as i and u (following a convention of Sanskrit grammar, where these
two vowels are the short substitutes of ē and ō), and occasionally a short e is even written
as a (when it is not preceded by the vowels a or ā).

pāṇḍavāḥ paṇḍavā
tāṁ taṁ
dēvīṁ dēviṁ
muhūrtaṁ muhuttaṁ
bhārgavāḥ bhaggavā

||3||
 One important context in which long vowels are shortened is when a word ends with a long
vowel and the following word is an enclitic particle that begins with a geminate consonant,
such as vva (= iva), ccia (= ēva).

Note that when the letters ō and ē are shortened in such a context, they are either written with
the same letters (i.e., there is no orthographic evidence of shortening), or they are written
with a short a.

putraḥ + ēva putto ccia

priyā + iva pia vva
muhūrtē + iva muhutte vva
§2.2.Deleting consonants
There will rarely be more than one consonant in the coda of a syllable, but in case there is, all
of the consonants apart from the one closest to the nucleus will be deleted:

paṅktiḥ (paṅk.tiḥ) paṁtī

P H O N EM E S
Syllable Structure
Lenition

§3. Introduction.

Vowels
The essential thing to know about the phonemic inventory of Prakrit is that several contrasts
have been eliminated. This not exactly the same as one sound “replacing” a set of sounds: as
we will see below, Prakrit only has a single sibilant, which is written as s, but that does not
mean that it was pronounced as a dental s in all contexts; it only means that there is no longer
a phonemic distinction between dental, retroflex, and palatal sibilants.

§4. Vowels.

Introduction
Consonants
Prakrit’s vowels are:

a ā i ī u ū ē ō
The vowels of Sanskrit that have been eliminated are the syllabic r̥ and r̥̥̄ (as well as l̥ and l̥ ̥̄ ),
as well as the “long diphthongs” ai and au. The latter violate the “no superheavy
syllables” rules, since they are composed on the segmental level by a bimoraic ā followed
by i/ī or u/ū. They are replaced in almost all occurrences by ē and ō:

Sanskrit Prakrit
kaurava- kōrava-
In a few cases, they are replaced by the disyllabic sequence aï and aü.

§5. Consonants.

Vowels

||4||
 §5.1.Semivowels
All of the semivowels of Sanskrit appear in Prakrit as well. Generally, the
semivowels r and l remain unchanged, except in conjunct consonants (see below); the
semivowel v also remains unchanged in many cases, except between a-vowels, where it is
usually lost (see below).

The exception is the semivowel y. In word-initial position it becomes the palatal stop j. There
are three possible contexts where it can occur, and it is lost or changed in all of them:

Sanskrit Prakrit
Example
Context Outcome
Sanskrit Prakrit
#yV- (word-initial) j- yā jā
-VyV- (single intervocalic) -∅- haya- haa-
-Cy- (conjunct) -∅- vyasana- vasaṇa-
§5.2.Sibilants
Prakrit only has a single sibilant, s, which is used in place of the three sibilants s, ś, and ṣ in
Sanskrit.

Sanskrit Prakrit
śara- sara
kaṣāya- kasāa-
§5.3.Nasals
Prakrit has two nasal phonemes, ṇ and m. The distinction between dental and retroflex nasals,
in other words, has been erased. (Even in Sanskrit, velar and palatal nasals are generally only
used at the end of a syllable, before velar and nasal stops, respectively, and hence they are not
really distinct phonemes. They never occur at the beginning of a syllable in Prakrit.)

At the end of a syllable, any nasal may be represented either by anusvāraḥ or by the class
nasal of a following stop consonant (parasavarṇaḥ), depending on the orthographic
preferences of the scribe/editor.

Especially in Jain manuscripts (i.e., manuscripts copied by Jain scribes) both the dental and
retroflex nasals are used, but in complementary distribution. That is:

 n is used at the beginning of a word (e.g., na);

 ṇ is used within a word (e.g., vaṇaṁ), except that
 a geminate nasal is written as nn rather than ṇṇ (e.g., rannaṁ for araṇyam).

L E N IT I O N
Phonemes
Conjuncts

§6. Introduction.

Voicing

||5||
 Lenition refers to the weakening of sounds in certain contexts. There are three types of
lenition in Prakrit, all of which apply to stop consonants:

 voicing of voiceless (retroflex) stops;

 debuccalization of aspirated stops (referring to the loss of place features); and
 loss lōpaḥ of other stops.

Lenition takes place only when the consonant occurs between two vowels. This implies that
the consonant is not part of a conjunct. Generally conjunct consonants are not subject to
lenition.

Another important aspect of lenition is that it only takes place within a phonological word.
That is, a single consonant standing at the beginning of one word, preceded by the final
vowel of the previous word, is not subject to lenition. This is a useful tool for identifying
word boundaries in Prakrit: a single intervocalic stop consonant, if it is not weakened, will
almost always be the initial consonant of a word.

Words preceded by the proclitic particle ṇa form a phonological word with that particle, and
hence a word-initial consonant can be recategorized as ‘between two vowels’ when the
negative particle precedes. We therefore have outcomes such as:

na jānānti ṇa āṇaṁti
na jānīmaḥ ṇa āṇimō

§7. Voicing.

Introduction
Loss
The voiceless retroflex stops ṭ and ṭh are voiced into ḍ and ḍh when they are single and
intervocalic:

ghaṭa- ghaḍa-
śaṭha- saḍha-
The voiced retroflex stops ḍ and ḍh remain as they are in single intervocalic position:

bāḍham bāḍhaṁ
vāḍava- vāḍaa-
Note that there are a few words that contain a voiceless dental stop in Sanskrit, but a retroflex
stop in Prakrit. In these cases we can usually assume that the stop has been retroflexed (e.g.,
by the r in prathama-) and the retroflex stop is then voiced:

prathama- paḍhama-
*śr̥thila- siḍhila-
Note that the Sanskrit word śithila- is probably a back-formation from siḍhila-, which is
certainly derived from the verbal root śrath “to be slack.”

In certain contexts, namely between a-vowels, the stop consonant p becomes a voiced
semivowel:

||6||
 pāpa- pāva-
prāpnōti pāvaï
The voiced stop b obviously cannot be voiced, but it is usually weakened into a semivowel
when it comes in between vowels:

anubandha- aṇuvaṁdha-

§8. Loss.

Voicing
Hiatus-fillers
The consonants k, g, c, j, t, and d are lost in intervocalic position. In their place a hiatus
stands between two vowels. Sometimes, however, a hiatus-filling consonant is introduced
instead. The forms illustrated here maintain the hiatus.

stop Sanskrit Prakrit

k makara- maara-
g mr̥ga- maa-
c kaca- kaa-
j gaja- gaa-
t gata- gaa-
d mada- maa-
The consonant y also undergoes loss in single intervocalic position, but because of the use
of y as a hiatus-filler, it sometimes looks as if it has not disappeared:

Sanskrit Prakrit
viyōga- viōa-
dayālu- daālu-
nayana- ṇaaṇa-
The consonant v is also supposed to be lost in this context, but it is sometimes retained:

Sanskrit Prakrit
lāvaṇya- lāaṇṇa-
Finally, the sound p is also included in the list of consonants that are lost whe single and
intervocalic, but it appears that this change is only regular when the adjoining vowels
are not a or ā. In that case, p is not lost, but becomes the voiced semivowel v (see the section
on voicing below).

Sanskrit Prakrit
ripu- riu-
supuruṣa- suurisa-
sapatnī- savattī-

||7||
§9. Hiatus-fillers.

Loss
Debuccalization
As a result of the widespread loss of single intervocalic stop consonants, Prakrit very often
has a hiatus between two vowels, which is generally not tolerated in Sanskrit. It appears that
the earliest grammars allowed this hiatus to stand. But in many manuscripts, the hiatus
between two vowels is interrupted by what Hemacandra, in his Prakrit grammar (8.1.140),
calls a “lightly pronounced y-sound” laghuprayatnatarayakāraśrutiḥ. This y-sound is properly
found only between two a-vowels (i.e., a or ā). Hence it is possible to write ṇayaṇaṁ instead
of ṇaaṇaṁ (for nayanam), gayā instead of gaā (for gatāḥ), and son on.

§10. Debuccalization.

Hiatus-fillers
All aspirate stops, except for ṭh and ḍh, are debuccalized when they occur between vowels:
that is, they lose their place of articulation, and only the aspiration remains, in the form of the
consonant h:

Sanskrit Prakrit
makha- maha-
mēgha- mēha-
grathita- gahia-
vadha- vaha-
śēpha- sēha-
ābhā- āhā-
There are no examples of ch and jh, because these sounds always occur in a conjunct.

C O N JU NC TS
Lenition

§11. Introduction.

No complex conjuncts
It is with respect to inherited conjunct consonants—two or more consonants without any
vowel intervening between them—that Prakrit, and the Middle Indic languages more
generally, differ most from Sanskrit. There are two constraints to be kept in mind: no
complex conjuncts are allowed, and further, no complex onsets are allowed; what the word
‘complex’ means will be explained below. Together with the aforementioned constraint
on superheavy syllables, these constraints means that conjunct consonants in Prakrit
will always take one of two forms:

1. C₁.C₁(h): that is, a geminate (double) consonant that is ambisyllabic (the final [coda]
segment of one syllable is identical with the initial [onset] segment of the syllable); as
in Sanskrit, the geminate consonant may be aspirated, in which case the aspiration
only appears on the final segment.
2. ṁ-C: that is, a syllable-final nasal (which is not specified for place of articulation)
followed by a syllable-initial consonant.

||8||
 Both of these structures are homorganic, in that there is only one place of
articulation specified for the entire conjunct. Any other kind of conjunct must be converted
by one of the repair strategies discussed below (i.e., assimilation or anaptyxis) to one of these
two canonical structures.

§12. No complex conjuncts.

Introduction
No complex consets
Prakrit, like most Middle Indic languages, does not allow complex conjuncts. I use the word
‘complex’ to refer to a conjunct in which the two consonantal segments have
different places of articulation or different manners of articulation, or both. That is, in
addition to disallowing any conjunct made of up heterorganic consonants, it disallows the
following combinations:

 stop + semivowel
 stop + nasal
 sibilant + semivowel
 sibilant + nasal
 sibilant + stop

The one combination that is allowed to stand is a syllable-final nasal followed by a stop,
semivowel, or sibilant.

§13. No complex consets.

No complex conjuncts
Assimilation
We have seen that the language generally resolves complex conjuncts to simple conjuncts,
that is, conjuncts of the form -ṁ.C- or -C₁.C₁-. These results, however, are ambisyllabic, in
the sense that they belong to two syllables. This is because Prakrit observes a constraint
according to which no more than one consonant can stand at the onset of a syllable. If the
conjunct occurs at the beginning of a word, it will be reduced not just to a simple conjunct,
but to a single consonant.

However, if the word with an initial conjunct is used in a compound, there is an opportunity
for the ‘hidden’ segment of the conjunct (the remainder or śēṣaḥ) to resurface. In this case,
the word can optionally begin with a geminate consonant:

Sanskrit Prakrit (no gemination) Prakrit (gemination)

su-prāsa- su-pāsa- su-ppāsa-
sa-vraṇa- sa-vaṇa- sa-vvaṇa-
bahu-tyāga- bahu-cāa- bahu-ccāa-
See the examples of assimilation and anaptyxis below for further examples of how consonant
conjuncts are resolved at the beginning of a word.

§14. Assimilation.

No complex consets
Anaptyxis

||9||
 Assimilation is the most common strategy for dealing with complex conjuncts. In
assimilation, the place and manner of articulation features of one of the segments spreads to
the other. What results is generally a geminate consonant, that is, a consonant that takes up
two segmental ‘slots,’ but which has the same place and manner of articulation in both slots.

The major question in assimilation is: which consonant wins out over the other? To answer
this question, we must first take into account the sonority of the consonant, and then
its position. Regarding sonority, the Jayadhavalā quotes the following verse:

dīsaṁti doṇṇi vaṇṇā saṁjuttā aha va tiṇṇi cattāri

tāṇaṁ duvvala-lōvaṁ kāūṇa kamō pajuttavvō

When two, or three or four, consonants appear in combination, elide the weakest one, and
continue the process.

This principle holds good, although it must be noted that a segmental ‘trace’ (called a śēṣaḥ)
of the elided consonant remains, and that it is a general principle which admits of many
exceptions. The notion of ‘weakness’ requires some explanation. It probably refers
to sonority, according to which consonants are arranged in the following categories:

 vowels (weakest)
 semivowels: y, v, r, l
 sibilants: ś, ṣ, s, h
 nasals: n, ṇ, m
 stops (strongest)

(Generally nasals are considered more sonorous than sibilants, but for the purpose of Prakrit’s
assimilation rules, they are “stronger” than sibilants.)

Among the semivowels, the strongest is l, followed by v, r, and y, in that order, and hence the
outcomes are:

Sanskrit Prakrit
kartavya- kattavva-
kalyāna- kallāṇa-
vraṇa- vaṇa-
kāvya- kavva-
sarva- savva-
The semivowel y, being relatively weak, is usually assimilated to an adjacent nasal, sibilant,
or stop. But when it comes after a dental stop, it palatalizes the stop before disappearing.
Hence:
Sanskrit Prakrit
anya- aṇṇa-
asya assa
avaśya- avassa-
tyāga- cāa-
dhyāna- jhāṇa-
pratyaya- paccaa-

||10||
Similarly, the semivowel r is also lost in most combinations, but when it follows a dental
stop, it often has the effect of retroflexing the stop before disappearing:

Sanskrit Prakrit
ārya- ajja-
murmura- mummura-
arṇava- aṇṇava-
sraj- saā-
śravaṇa- savaṇa-
krānta- kaṁta-
arka- akka-
grasta- gattha-
vartatē vaṭṭaï
vardhatē vaḍḍhaï
trāsa- tāsa-
citra- citta-
In combination with a sibilant, r is usually lost, although often with lengthening of the
preceding syllable; in some cases, the conjunct -rś- becomes -ṁs-:

Sanskrit Prakrit
īrṣyā- īsā-
miśra- mīsa-
ādarśa- ādāsa-
darśana- daṁsaṇa-
The semivowel v can sometimes labialize an adjoining stop before it disappears:

Sanskrit Prakrit
dvādaśa- bāraha-
dviguṇa- biuṇa-
dvāra- bāra-
The sibilants are lost when they adjoin nasals and stops, but only after transferring aspiration
to the adjoining consonant. Sibilants are therefore an important source of aspirate consonants
in Prakrit, including aspirated nasals, which do not exist in Sanskrit.

Sanskrit Prakrit
snāna- nhāṇa-
śmaśāna- mhasāṇa-
vismaya- vihmaa-
stuti- thuī-
āścarya- accharia-, acchēra-
suṣṭhu suṭṭhu
śāstra- sattha-
Note that the s of the prefixes nis and dus does not cause aspiration: niccala- from niścala-
, duccaria- from duścarita-.

||11||
When s follows a dental consonant, besides aspirating it, it also palatalizes it. Thus:

Sanskrit Prakrit
vatsa- vaccha-
matsya- maccha-
The conjunct kṣ needs to be mentioned separately, because even though by the above rules it
should develop into (k)kh, in a few cases it develops into (c)cch. But there is widespread
disagreement about what the ‘correct’ outcome should be. Generally we
encounter khaṇa- for kṣaṇa-, and acchi- for akṣi-, but sometimes also chaṇa- and akkhi-. The
word lacchī (lakṣmī) is always spelled thus.

The paradigm of as “to be,” which is enclitic in Prakrit and loses its initial vowel, serves as a
good example of the assimilation of sibilants (Sanskrit forms are on the left, and Prakrit
forms on the right):

ēkavacanam bahuvacanam
prathamapuruṣaḥ asti tthi santi saṁti
madhyamapuruṣaḥ asi si stha ttha
uttamapuruṣaḥ asmi mhi smaḥ mhō
Nasals are generally lost in combination with stops:

Sanskrit Prakrit
agni- aggi-
bhagna- bhagga-
However, there are a few cases in which the nasal is assimilated to the stop only after
transferring its place of articulation properties to the stop. That is, a m can labialize an
adjoining k or t, as in:

Sanskrit Prakrit
ātma- appa-
rukminī- ruppiṇī-
Moreover, there are cases in which the constraint is satisfied by anaptyxis (see below).

If two conjuncts that have the same sonority remain—that is, two semivowels, or two nasals,
or two stops—then regressive assimilation is the usual outcome, i.e., the first consonant is
assimilated to the place of articulation of the second.:

Sanskrit Prakrit
janma- jamma-
bhukta- bhutta-
utkaṇṭha- ukkaṁṭha-
śabda- sadda-
There are, however, some exceptions, in which progressive assimilation seems to occur:

mukta- mukka-

||12||
§15. Anaptyxis.

Assimilation
Anaptyxis, or the insertion of a vowel in between two consonants, is a much rarer strategy of
avoiding complex conjuncts. It is sometimes used in words involving the conjuncts -tn- or -
dm-, as well as in words that begin with a combination of a sibilant and a nasal or semivowel.

The vowel that is inserted is a by default, but in case one of the consonants is a labial, it is u.
In a few cases, it is i. The anaptyptic vowel feeds the process of intervocalic vowel lenition,
and hence we have outcomes such as:

Sanskrit Prakrit
ratna- raaṇa-
padma- paüma
ślāgha- silāha-
snēha- siṇēha-
snāna- siṇāṇa-
Note that anaptyxis in conjuncts involving sibilants and nasals is optional; in other words, it
is also possible to assimilate the sibilant to the nasal, either with aspiration or without,
e.g., nēha- for snēha-, ṇhāṇ- for snāna-, etc.

||13||